OVERVIEW
====================================================================================================
Replication code for

Finkelstein, Notowidigdo. Take-up and Targeting: Experimental Evidence from SNAP.

Production date: January 29, 2019

DATA ACCESS: RESTRICTED DATA
====================================================================================================
The code in this archive requires access to restricted administrative data. The code is designed to 
access these data via a Subversion repository we designed for this project and is used to produce the 
set of empirical results present in the paper from a cleaned dataset, rather than completely raw data. 
Researchers wishing to replicate our results will need to obtain their own copy of the data and clean 
the data as is described in the paper.

DIRECTORY STRUCTURE
====================================================================================================
Our code directory contains 11 do files that run analysis on a single cleaned dataset, and a variable
library that is used to define all the global variables used in the analysis code. We recommend running
analysis code in the following sequence in order to replicate the full set of results. We describe each
file and the analysis results it produces below.

DESCRIPTION OF EACH FILE  
====================================================================================================
predict_benefit_class	Generates Figure A6, and estimates predicted benefits using everyone in the 
			study population who enrolled in SNAP in the 9 months following the initial 
			mail date and for whom we observe a benefit amount. The process is described 
			in Section D in online appendix.

predict_enrollment	Estimates predicted benefits using everyone in the 
			study population who enrolled in SNAP in the 9 months following the initial 
			mail date and for whom we observe a benefit amount. The process is described 
			in Section D in online appendix.

f-stat			Estimates F-statistics in Table A3, Table A4 to check balance across treatment
			arms. The F-statistic (and associated p-value) is calculated based a regression 
			in which we “stack” all of the variable values into a single left-hand side 
			outcome variable and interact the treatment indicator with variable fixed effects

bos_analysis.do		Generates Table A19, Table A20. In addition to the main cleaned analysis data,
			it also uses the information on Benefit Outreach Specialists (BOS) that we
			separately received from Benefit Data Trust (BDT).

exhibits_appendix.do	Generates Table 6, Table A3, Table A4, Table A5, Table A6, Table A7, Table A8,
			Table A9, Table A10, Table A13, Table A14, Table A15, Table A16, Table A17.In 
			addition to the main cleaned analysis data, it also uses call data that we 
			separately received from BDT.

exhibits_compliers_AT_NT
exhibits_compliers_NT_C
exhibits_compliers	These three files together generate Table A7, Table A11, Table A12. The process
			of estimating means of compliers, always takers, and never takers, and of 
			estimating p-values is described in Section F in online appendix.

exhibits_main		Generates Table 1, Table 2, Table 3, Table 4, Table 5, Table A18.

exhibits_figures	Generates Figure 1, Figure A7, Figure A8.

lee_bounds		Generates Table A1. The process of estimating Lee bounds is described in Section 
			D in online appendix.